Keynote: Kubeflow ML on Kubernetes
summary
Announces Kubeflow 0.1/1.0 as a packaged, Kubernetes-native ML stack with Jupyter, TensorFlow training, TensorBoard and model serving; demo shows a local Minikube → GKE with GPUs/TPUs workflow where training pods and TPU jobs run and a prebuilt model serves real-time sentiment results, though full training can take ~1 hour so a pre-trained model is used for the demo. Claims composability, portability and scalability, cites 70+ contributors, 17 companies, and says components are usable without forking the Kubernetes API.
AI-written summary · about
chapters (13)
- 0:00Start
- 0:23I'm Dave, he's Vish
- 3:34Kubernetes extensibility
- 4:25Composability
- 5:18Portability
- 5:53Scalability
- 6:36Oh, you want to use ML on K8s?
- 7:50What's in the box?
- 16:56Reminder: what you didn't see
- 18:34It's a Whole New World
- 18:59Using Data to Save Lives
- 20:13We're just getting started!
- 21:27Kubeflow is open!
related talks
15mKeynote: Machine Learning on Kubernetes Made Easy With Kubeflow
37mTowards Kubeflow 1.0, Bringing a Cloud Native Platform For ML to Kubernetes
14mKubeflow: Where ML Meets Kubernetes
10mKeynote: Machine Learning using Kubeflow and Kubernetes
21mIntroducing Kubeflow (w. Special Guests Tensorflow and Apache Spark)
12mKeynote: Building Enterprise AI/ML Platform on Kubernetes
46mKubeflow, TensorFlow on Kubernetes
46mDeploying scalable Machine Learning models with Kubernetes